Side swipe accidents occur primarily when drivers attempt an improper lane change, drift out of lane, or the vehicle loses lateral traction. In this paper, a fusion approach is introduced that utilizes multiple differing modality data, such as video data, GPS data, wheel odometry data, potentially IMU data collected from data logging device (DL1 MK3) for detecting driver’s behavior of lane changing by using a novel dimensionality reduction model, collaborative representation optimized projection classifier (CROPC). The criterion of CROPC is maximizing the collaborative representation based between-class scatter and minimizing the collaborative representation based within-class scatter in the transformed space simultaneously. For lane change detection, both feature-level fusion and decision-level fusion are considered. In the feature-level fusion, features generated from multiple differing modality data are merged before classification while in the decision-level fusion, an improved Dempster–Shafer theory based on correlation coefficient, DST-CC is presented to combine the classification outcomes from two classifiers, each corresponding to one kind of the data. The results indicate that the introduced fusion approach using a CROPC performs significantly better in terms of detection accuracy, in comparison to other state-of-the-art classifiers.

中文翻译：

---

使用多传感器数据融合个性化检测换道行为

侧滑事故主要发生在驾驶员尝试不正确的换道、偏离车道或车辆失去横向牵引力时。在本文中，介绍了一种融合方法，该方法利用多种不同的模态数据，例如视频数据、GPS 数据、车轮里程计数据、从数据记录设备 (DL1 MK3) 收集的潜在 IMU 数据，通过使用新颖的降维模型，协同表示优化投影分类器（CROPC）。CROPC 的准则是最大化基于类间散布的协同表示，同时最小化基于类内散布的协同表示。对于车道变换检测，同时考虑特征级融合和决策级融合。在特征级融合中，由多种不同模态数据生成的特征在分类前进行合并，而在决策级融合中，提出了一种基于相关系数的改进的 Dempster-Shafer 理论，DST-CC 将两个分类器的分类结果结合起来，每个分类器对应一种数据。结果表明，与其他最先进的分类器相比，引入的使用 CROPC 的融合方法在检测精度方面的表现要好得多。